Why Perception-Response Time (PRT) Is Not Like Gravity
It is sometimes easier to explain a concept by saying what it is not, e.g., seeing is not a homunculus in the head looking at a screen. Similarly, it is useful to show that PRT is not by nature like gravity, especially since the previous chapter unintentionally reinforced the errant impression that PRT is exactly like gravity.
For present purposes, gravity has three important properties. First, it is a variable, but it can reasonably be treated as a constant. Its assumed value is always 32.2 ft/sec2, even though, strictly speaking, its exact value depends on several factors. In most applications (on earth), the error is too small to matter. Since it is a virtual constant, it may be applied in a cookbook manner with no need to understand the origin, history, effects of various circumstances or underlying scientific basis. Second, a value for gravity is always necessary when performing braking and similar accident reconstruction calculations. It is not possible to simply say that the value is too uncertain to determine, to use qualitative terms such as more or less acceleration or to deem it irrelevant.
A third difference is that gravity has no variability. It is a simple, single number. Humans have an innate urge to reduce cognitive complexity, i.e., find a simple answer to a complex problem. It would be nice if collisions could be analyzed with a single number. There is a certain amount of time to avoid collision and either the driver's PRT is longer or shorter than this time. If it is shorter, he avoids collision. If it is longer, he does not. What could be simpler?
It is perhaps natural that accident reconstructionists and engineers, people trained in the physical sciences, should treat PRT like a cookbook physical constant. They need gravity to compute braking as well as to perform many other physical analyses. They also want a simple number that they can use without any deep understanding. Psychology and behavior is not their business and they don't want to be diverted into the extensive time and effort required to obtain a real understanding of the phenomenon. To them, it must seem a small step to cross over into human factors and to analyze avoidability where they will need a PRT value just as they needed a gravity value to perform the purely physical analysis.
It is not surprising then that they view PRT as a human factors analog to gravity - a fixed number where variability, context and scientific basis can be ignored. If AASHTO says that PRT is 2.5 seconds, then there is no need to bother reading the Green Book in order to learn the method that AASHTO used to determine the value, the assumptions they made or what they intended it to represent. There is no need to read the original studies upon which AASHTO relied in reaching their conclusions. In fact, learning about such issues is critical to anyone attempting to apply this value to the real-world. I explain this in depth later when I discuss how the AASHTO derived their values and the problems with their choice of base data.
Unfortunately, PRT is not like gravity at all, and crossing from physics to human factors, i.e., experimental psychology, is a huge step. Psychology is far more complex1 than physics , and PRT is a far more complicated concept than gravity. It is not reasonable to treat PRT as a constant, or even a small set of constants, that apply universally across all situations. Mean/median PRT varies wildly due to circumstances and PRT distributions often have very large variability and large skew. The number of factors affecting PRT is far greater and more difficult to quantify than the factors that determine gravity. In sum, to assign a value for gravity, it is not necessary to be a physicist or to have read the underlying physics research. To assign a PRT value to a real situation, or even know whether it is feasible to assign a number, it is absolutely necessary to know a great amount of the underlying research in both PRT and in behavioral psychology in general.
Sometimes, the key question is not the speed of PRT but whether PRT is even a relevant issue. Such an assertion may seem unintuitive since a driver response is required for avoidance. This is again thinking in terms of the gravity analogy. In reality, there are several reasons for ignoring PRT in many cases.
1. The relationship between PRT and accident causation is weak.
One study (Muttart, 2005) found no relationship between PRT and accident involvement while another (Mihal & Barret, 1976) found that individual differences in PRT had no relationship to accident rate. A third (Ayres, & Kubose, 2012) concluded that for a given TTC, PRT often failed to distinguish between those who crashed and those who did not. Further, Malaterre, Ferrandez, Fleury & Lechner (1988) concluded that under extreme emergency, instinctive reflexes take over so that all drivers "become equal". In fact, most accidents involve drivers who have good driving records and who have not had a previous major crash (Campbell, 1959). The most likely explanation for such results is that situational factors, speed, time, distance, light and contrast, and not PRT, primarily determine whether an accident will occur. Whether a driver can avoid a collision is often due to plain old luck, i .e., it "is "rather like taking a bet" (Prynne & Martin, 1995) and "therefore a matter of chance combination of circumstances" (Baker, 1960). In sum, there is little evidence that accidents occur because there are "bad apples" who respond abnormally slowly.
Many will find this conclusion difficult to accept because it runs contrary to a strong human cognitive bias, "fundamental attribution error" (Ross, 1997). When a person judges the cause of an event, he can assign it to either dispositional factors inside the person or to situational factors outside the person's control. People are heavily biased toward blaming individual disposition even though most people act the same way in the same situation. Fundamental attribution error is very powerful and highly resistant even to strong evidence that environmental constraints were the primary cause. It is also one of the prime promoters of hindsight bias.
2. Long PRT is often a side effect, not the cause, of avoidance failure.
That is, the cause is the factors that resulted in a PRT that was too long, and not the PRT itself. Think about it this way. Assume the fastest possible driver PRT is 1.5 seconds. The driver responds in 2.5 seconds on a dark road at night to a pedestrian. If there were research studies that would contain PRT data for a pedestrian in his specific clothing, on his specific location on the roadway, with the specific street lighting etc. then that might be used directly. Unfortunately there are no such data and likely never will be. At this point, there are only a few ways to proceed. One is to make up a number, which in my experience is all too common. ("Well, the standard PRT is 1.5 seconds in daylight, so I added another second for nighttime"). Another is to use an illuminance criterion, such as the 3.2 lux twilight value. This is another cognitive complexity reducer since it is simple and requires no understanding. The chapter on contrast detection has already explained why this method is unreliable.
So what is the alternative? Perhaps the best strategy is to shift away from PRT to the factors that determine PRT. After all, response is just the final event in a chain of human information processing, i.e., sensing, identification, situational awareness and response selection. It is an effect of these factors and only a manifestation of them. To see this, consider how PRT might be determined. Under absolutely ideal conditions, humans can reliably respond to an unexpected road event in roughly 1.0-1.5 second, depending on whom you ask. This sets the absolute limit on what a driver can achieve. If the available time is less than this, then avoidance is simply impossible. If it is longer, then inevitable uncertainties arise. The real world is never ideal so the question is always, how much more is the PRT going to be. There are many factors that may inflate the PRT: visibility, conspicuity, expectation violation, complexity, novelty, weather, etc. The amount by which each of these factors individually, let alone in groups, raise PRT is generally difficult to say with much certainty. The important point is that the issue is not the driver's PRT but rather the conditions that caused the increase to whatever PRT that driver actually produced.
The real performance determining factors should be examined, and not the PRT that they produce. If lighting and contrast are low, then PRT will be long. If the driver fails to see the object in time to avoid because of lighting conditions, the issue is visibility, not PRT (at least directly). If the driver fails to see the visible object in time because other road objects attracted his attention, or because he is texting on his smart phone, the focus should be the events that controlled attention, not PRT. Certainly, a specific number like 2.0 seconds results in a nice fuzzy feeling and the pretence of scientific precision, but in most cases this is merely illusory confidence.
In sum, the process should work backward from the way people normally think of it. The question accident investigators typically ask is "What is the expected PRT"? The better question is "What factors caused the driver's PRT, what ever it was, to be insufficient?" The task is to explain what actually happened and not to hypothesize some specific number in the absence of real scientific evidence. The fundamental cause of most collisions is likely "late detection" (Rumar, 1990). Anyone who has investigated collisions knows that drivers commonly say that they never saw the pedestrian, etc. or saw him just a fraction of a second before the collision. What is the point in considering PRT in such cases? The real issue is the reason that detection was late.
3. Variability is critical
Variability is irrelevant to applying a value for gravity but absolutely central to interpreting PRT. Gravity is always 32.2 ft/sec2, end of story. In PRT, variability is as critical as the measure of central tendency (mean/median), since it defines the range of normal behaviors. Research studies are little help for determining variability, beyond setting a lower limit. As I have explained, research studies are designed to minimize variability, so they typically underestimate the uncertainty of real-world behavior. This is most important in situations that require thinking, i.e., ones that are not highly reflexive.
Moreover, there is still the additional problem in defining normality even if variability could be convincingly determined. Suppose the mean PRT is 1.5 seconds and a driver responds in 2 seconds. Is this within the realm of normal driver behavior? The question is unanswerable without knowing variability. If the standard deviation is .25 second, then he is two standard deviations above the mean and is in the 5% of slowest drivers. If the standard deviation is .5 second, then he is only one standard deviation above the mean and is in the slowest 16% of drivers. Where are the limits of "normality?" The one or the two standard deviation above the mean driver? In contrast sensitivity, the Adrian contrast model sets the value at the 99.63% detection level (and then adds multipliers). That is more than three standard deviations above the mean and is a far looser definition of normality. Which definition is correct? Of course, PRT distribution exhibit strong kurtosis toward longer values, so standard deviations may greatly underestimate the number of the slow times.
There is one final problem in defining variability as well as mean/median. That is the problem of defining PRT itself. In counterfactual thinking, it is always easy to say that the driver could have avoided the collision if he had responded faster. This may be trivially true, but then the important questions are 1) what is meant by "responded"? and more importantly, 2) responded faster to what? When does the PRT clock start and when does it stop? This is not as simple or obvious as may appear at first glance, as I explain in the next section.
1Nobel Prize winning physicist Murray Gell-Mann once compared the relative complexities of physics and psychology when he observed, "Imagine how hard physics would be if particles could think." Michael Shermer in The Believing Brain notes that the term "hard science" is commonly used backwards. Physics is often called "hard" while psychology is called "soft." In fact, psychology is far "harder" than physics as explained. Psychology is the real "hard science".
This article is a preview of a version in the book Roadway Human Factors: From Science To Application (2017).